A descriptive pattern recognition system applied to pictorial patterns where the discriminating information is carried in the object shape

A descriptive classifier has been developed for silhouetted pictorial patterns. It has the capability of recognition with inherently high discriminatory power between an essentially unlimited number of discrete pattern classes. The user is allowed substantial latitude in determining what pictorial instances of objects should be admitted to the same pattern category with pattern class constraints than can be adjusted in an iterative fashion. The user can also generate and display random instances of a pictorial pattern class until, according to subjective evaluation, the pattern class defined within the machine is the same as that envisioned by the user. Pictorial classification is performed using a system-generated figure classification number. The figure classification number defines a unique point within a classification space having an assigned pattern class name. The pattern class can, by user selection, be made to include many or few unique figure classification states. The system´s performance is illustrated by classification of a series of pictorial silhouettes